University of Pennsylvania 11th annual conference on statistical issues in clinical trials: Estimands, missing data and sensitivity analysis (afternoon panel session).

Andrea B. Troxel,Eric J Tchetgen Tchetgen,Roderick J Little

doi:10.1177/1740774519853565

Abstract

University of Pennsylvania 11th annual conference on statistical issues in clinical trials: Estimands, missing data and sensitivity analysis (afternoon panel session).

Highlights

We know from many studies long ago that statins are effective in reducing cholesterol with relatively manageable side effects, and so it should be easy to just give patients statins, and they should do better
One is adherence, which was measured by electronic pill bottles so that we knew when the patients were opening their bottles, and the other was incidence of either a new prescription for statins or intensification of the statin, either increasing dose or a change to a more potent medicine
I don’t have a plot for the intensification or prescription activities, but we found, again, what in retrospect seems obvious: that in the shared incentive arm and in the physician incentive arm, there was a statistically significantly higher rate of intensification, and that did not occur in the patient incentive arm and the control arm

Summary

Introduction

Most incentives are imperfect, and so it turns out you could randomize incentives, and this could be done within the treatment arms of your randomized trial, and you have this new experimental design that’s targeting the missing data process, provided it satisfies the aforementioned untestable condition about independence between the incentive and the outcome Y in the population. This issue of defining adherence very precisely is something that I’ve given a lot of thought to in the context of behavioral trials (that I’ll describe in a minute) and I think that’s something we have to focus hard on.

Results

Conclusion